[SPARK-59173][ML] Use isEmpty/nonEmpty instead of size comparisons in MLlib

[uros-b]

What changes were proposed in this pull request?

This replaces size-vs-zero comparisons with the direct emptiness / non-emptiness predicates in the mllib module:

• Emptiness (x.size == 0 -> x.isEmpty): TreeEnsembleModel.totalImportances (an OpenHashMap, which extends Iterable) and two Seq[ShuffleDependency] checks in ALSSuite.
• Non-emptiness (x.size > 0 -> x.nonEmpty): DecisionTreeMetadata.featureArity and Strategy.categoricalFeaturesInfo, both Scala Map[Int, Int].

There are no Java-collection cases in this module.

Deliberately left unchanged, because they are not collection-emptiness checks -- they are mllib.linalg.Vector dimensions or numeric sizes/thresholds, and Vector has no isEmpty/nonEmpty:

• ChiSqTest (expected / observed are Vector dimensions, compared with != 0 and == 0.0);
• BinaryClassificationPMMLModelExport and GeneralizedLinearPMMLModelExport (model.weights, a Vector);
• MultivariateOnlineSummarizer (instance, a Vector -- the message reads "Vector should have dimension larger than zero");
• BlockMatrix (a matrix row Vector);
• ALS (RatingBlockBuilder.size, a custom builder with no nonEmpty);
• SpearmanCorrelation (cachedUids.size >= 10000000, a size threshold, not an emptiness check).

Why are the changes needed?

isEmpty / nonEmpty state the intent directly. Behavior is unchanged: every converted receiver (OpenHashMap, Seq, Scala Map) defines the predicate as exactly equivalent to the original comparison.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

Existing tests. This is a behavior-preserving refactor; the mllib module compiles cleanly. Counterpart of SPARK-59149 (SQL) and SPARK-59172 (core).

Was this patch authored or co-authored using generative AI tooling?

Generated-by: Claude Code (Opus 4.8)

[uros-b]

Waiting for CI.

[uros-b]

cc @dongjoon-hyun @HyukjinKwon here, regarding #58449.

[uros-b]

added @HyukjinKwon for ML